The Department of Defense's Chief Digital and Artificial Intelligence Office (CDAO) expanded its production Other Transaction Authority (OTA) agreement with Scale AI Inc. from $100 million to $500 million, the company announced, with the government citing demand across the department that had "exceeded the scope of the original contract." The expansion — a fivefold increase from the September 2025 award — reflects the speed at which AI capabilities are being integrated across DoD's operational and analytical workflows.

What Scale AI Does for the Pentagon

Scale AI's core business is data labeling and AI model training infrastructure — it provides the human and automated pipelines that turn raw data into training sets that machine learning models can learn from. For the Pentagon, this translates into four main service areas covered under the CDAO agreement:

Scale Data Engine handles computer vision and multimodal machine learning pipelines — labeling satellite imagery, drone footage, and sensor data to train recognition and targeting models used across intelligence analysis and logistics planning.

Scale GenAI Platform provides secure generative AI model deployment across DoD's classification tiers: NIPR (unclassified), SIPR (secret), and JWICS (top secret/SCI). This is a critical capability — most commercial large language model APIs are not cleared for classified use, so Scale's ability to deploy and manage models within DoD's classified enclaves fills a significant operational gap.

Scale Donovan is a generative AI decision-support tool designed for military operators — think of it as a classified AI assistant trained on defense-specific data and integrated with operational systems. It was originally developed under CDAO's Task Force Lima and has been piloted with multiple combatant commands.

Engineering capability development sprints allow DoD components to contract for rapid prototyping of custom AI capabilities, with Scale's engineering teams working alongside government data scientists to build mission-specific models on accelerated timelines.

OTA Structure and Ordering Mechanics

Production OTAs operate differently from standard federal contracts. DoD components that want to order Scale AI services don't need to run their own competition — they can initiate a Project Agreement under the CDAO umbrella agreement at pre-negotiated, volume-based pricing, with CDAO providing partial co-funding. This dramatically accelerates the timeline from identified need to signed agreement: weeks rather than months or years under traditional FAR-based procurement.

The trade-off is accountability: OTAs have fewer protest rights for competitors, less public transparency, and lighter audit requirements than FAR contracts. Critics in Congress have raised concerns about the proliferation of large OTAs for AI services, arguing that the speed advantage should not come at the cost of competition and oversight. The Scale AI expansion to $500 million will likely draw additional scrutiny given the size of the ceiling and the company's high political profile.

Meta's Stake and the Conflict-of-Interest Question

Meta Platforms holds a 49 percent non-voting stake in Scale AI, acquired in June 2025 in a deal that valued Scale at approximately $14.3 billion. The non-voting structure was specifically designed to address concerns about foreign influence and conflict of interest in Scale's government work — Meta's stake does not give it board seats or operational control, and Scale maintains a separate government entity with cleared personnel and systems.

Nevertheless, the arrangement has attracted attention from congressional oversight staff, who have asked CDAO to clarify how Scale AI's security protocols prevent Meta's data practices from influencing or accessing classified DoD AI workflows. CDAO has provided classified briefings to relevant committees; the public record does not reflect any conclusion that the arrangement violates existing conflict-of-interest standards.

The Anthropic Contrast

The Scale AI expansion comes as DoD navigates a high-profile dispute with Anthropic, whose Claude AI models were designated a "supply chain risk" after the company refused to allow its AI for autonomous weapons development and domestic surveillance use cases. A federal court has temporarily blocked DoD's directive to remove Anthropic products from government systems. The Scale AI relationship — framed explicitly as avoiding single-provider dependency — is partly a response to lessons learned from that dispute.

Under Secretary of Defense for Research and Engineering Emil Michael stated publicly on May 8 that the department would "never again be single-threaded with any one model" — a direct reference to the period when Microsoft's OpenAI-backed Azure was the dominant AI platform for most DoD components. The expanded Scale AI agreement, alongside separate OTAs with Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, Oracle, and SpaceX, reflects a deliberate multi-vendor AI strategy.

What It Means for Contractors

  • Firms that have not yet established relationships with CDAO's AI Rapid Capabilities Cell should do so immediately — the Scale OTA and peer agreements represent the primary acquisition pathway for AI services across DoD for the foreseeable future.
  • System integrators with cleared facilities and IL5/IL6 cloud environments are in high demand to deploy and operate Scale AI capabilities within classified networks; Scale's government entity partners with primes for classified deployment work.
  • Small businesses specializing in AI data annotation, model validation, and responsible AI testing should explore Scale's supplier ecosystem — Scale contracts out significant volumes of human annotation work to specialized subcontractors.
  • Firms competing with Scale AI on future AI services acquisitions should be aware that CDAO's preference for multi-provider redundancy may create room for specialized niche players, but broad platform competition against an OTA incumbent at this scale is extremely difficult without an equivalent cleared infrastructure.

Sources